diff mbox

ARM: zImage: Allow DTB to override broken ATAG_MEM

Message ID 1399439776-18535-1-git-send-email-bjorn.andersson@sonymobile.com (mailing list archive)
State New, archived
Headers show

Commit Message

Bjorn Andersson May 7, 2014, 5:16 a.m. UTC
Support overriding ATAG_MEM, by specifying non-zero content of the /memory/reg
property in the appended DTB. This is needed to work around bootloaders passing
broken tags.

Signed-off-by: Bjorn Andersson <bjorn.andersson@sonymobile.com>
---
 arch/arm/boot/compressed/atags_to_fdt.c |   18 +++++++++++++++++-
 1 file changed, 17 insertions(+), 1 deletion(-)

Comments

Bjorn Andersson May 7, 2014, 5:32 a.m. UTC | #1
On Tue, May 6, 2014 at 10:16 PM, Bjorn Andersson
<bjorn.andersson@sonymobile.com> wrote:
[...]
> +static int fdt_overrides_atag_mem(void *fdt)
> +{
> +       const char *memory;
> +       int len = 0;
> +
> +       memory = getprop(fdt, "/memory", "reg", &len);
> +       if (memory) {
> +               while (len--) {
> +                       if (*memory != '\0')

Sorry, this is of course supposed to be:
                       if (*memory++ != '\0')

> +                               return 1;
> +               }
> +       }
> +
> +       return 0;
> +}

Regards,
Bjorn
Uwe Kleine-König May 7, 2014, 8:06 a.m. UTC | #2
On Tue, May 06, 2014 at 10:16:16PM -0700, Bjorn Andersson wrote:
> Support overriding ATAG_MEM, by specifying non-zero content of the /memory/reg
> property in the appended DTB. This is needed to work around bootloaders passing
> broken tags.
This feels wrong. I think it's quite usual that the device tree
specifies a non-0 /memory/reg property. I checked four more or less
random dts files[1], and three of them have this property set with
actual values.

So I wouldn't be surprised if this patch results in more damage than
it's worth. The optimal fix would be to make the bootloader do the right
thing. And if you trust your dtb more than your bootloader, disable
ARM_ATAG_DTB_COMPAT.

Best regards
Uwe

[1]
arch/arm/boot/dts/am335x-bone-common.dtsi
arch/arm/boot/dts/exynos5440-sd5v1.dts
arch/arm/boot/dts/imx6q-wandboard.dts
arch/arm/boot/dts/tegra30-cardhu.dtsi
> 
> Signed-off-by: Bjorn Andersson <bjorn.andersson@sonymobile.com>
> ---
>  arch/arm/boot/compressed/atags_to_fdt.c |   18 +++++++++++++++++-
>  1 file changed, 17 insertions(+), 1 deletion(-)
> 
> diff --git a/arch/arm/boot/compressed/atags_to_fdt.c b/arch/arm/boot/compressed/atags_to_fdt.c
> index d1153c8..b534cd6 100644
> --- a/arch/arm/boot/compressed/atags_to_fdt.c
> +++ b/arch/arm/boot/compressed/atags_to_fdt.c
> @@ -95,6 +95,22 @@ static void merge_fdt_bootargs(void *fdt, const char *fdt_cmdline)
>  	setprop_string(fdt, "/chosen", "bootargs", cmdline);
>  }
>  
> +static int fdt_overrides_atag_mem(void *fdt)
> +{
> +	const char *memory;
> +	int len = 0;
> +
> +	memory = getprop(fdt, "/memory", "reg", &len);
> +	if (memory) {
> +		while (len--) {
> +			if (*memory++ != '\0')
[added fixup noted in follow-up-mail]

> +				return 1;
> +		}
> +	}
> +
> +	return 0;
> +}
> +
>  /*
>   * Convert and fold provided ATAGs into the provided FDT.
>   *
> @@ -180,7 +196,7 @@ int atags_to_fdt(void *atag_list, void *fdt, int total_space)
>  		}
>  	}
>  
> -	if (memcount) {
> +	if (memcount && !fdt_overrides_atag_mem(fdt)) {
>  		setprop(fdt, "/memory", "reg", mem_reg_property,
>  			4 * memcount * memsize);
>  	}
> -- 
> 1.7.9.5
> 
> 
> _______________________________________________
> linux-arm-kernel mailing list
> linux-arm-kernel@lists.infradead.org
> http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
>
Bjorn Andersson May 7, 2014, 3:29 p.m. UTC | #3
On Wed, May 7, 2014 at 1:06 AM, Uwe Kleine-König
<u.kleine-koenig@pengutronix.de> wrote:
> On Tue, May 06, 2014 at 10:16:16PM -0700, Bjorn Andersson wrote:
>> Support overriding ATAG_MEM, by specifying non-zero content of the /memory/reg
>> property in the appended DTB. This is needed to work around bootloaders passing
>> broken tags.
> This feels wrong. I think it's quite usual that the device tree
> specifies a non-0 /memory/reg property. I checked four more or less
> random dts files[1], and three of them have this property set with
> actual values.

I thought u-boot did something like this, but after checking the code
it seems that I
was wrong.

But if that's not the case then you're right; we have to continue to
be bug-compatible
with all those dtbs out there.

>
> So I wouldn't be surprised if this patch results in more damage than
> it's worth. The optimal fix would be to make the bootloader do the right
> thing. And if you trust your dtb more than your bootloader, disable
> ARM_ATAG_DTB_COMPAT.

I unfortunately have a boot loader passing information in ATAG_CMDLINE that I
need, so I can't disable ARM_ATAG_DTB_COMPAT.


My problem is that the boot loader on every shipped Qualcomm MSM8x60, MSM8960
and APQ8064 based device passes ATAG_MEM with the incorrect start address. There
is no way to update the boot loader for these.

For development I have a .init_meminfo in my board file patching the
meminfo; this looks
really bad so I'm hoping we can find some alternative solution.

The proposed solution from some of the people working on this is to
(post build) patch the
zImage to inject code that corrects the ATAG_MEM before jumping to the
kernel; but I am
hoping we can find some sane way instead...

Regards,
Bjorn
Andreas Färber May 7, 2014, 9:24 p.m. UTC | #4
Am 07.05.2014 17:29, schrieb Bjorn Andersson:
> My problem is that the boot loader on every shipped Qualcomm MSM8x60, MSM8960
> and APQ8064 based device passes ATAG_MEM with the incorrect start address. There
> is no way to update the boot loader for these.

Chaining bootloaders comes to mind. This is being done on the Exynos5
based Chromebooks, for instance.

Regards,
Andreas
Pavel Machek June 14, 2014, 8:33 p.m. UTC | #5
On Wed 2014-05-07 23:24:18, Andreas F?rber wrote:
> Am 07.05.2014 17:29, schrieb Bjorn Andersson:
> > My problem is that the boot loader on every shipped Qualcomm MSM8x60, MSM8960
> > and APQ8064 based device passes ATAG_MEM with the incorrect start address. There
> > is no way to update the boot loader for these.
> 
> Chaining bootloaders comes to mind. This is being done on the Exynos5
> based Chromebooks, for instance.

You can easily start fixed u-boot from broken u-boot...

But I'm not sure that's the way to go. Perhaps your localised/ugly hack in board
init code is the way to go...
									Pavel
diff mbox

Patch

diff --git a/arch/arm/boot/compressed/atags_to_fdt.c b/arch/arm/boot/compressed/atags_to_fdt.c
index d1153c8..b534cd6 100644
--- a/arch/arm/boot/compressed/atags_to_fdt.c
+++ b/arch/arm/boot/compressed/atags_to_fdt.c
@@ -95,6 +95,22 @@  static void merge_fdt_bootargs(void *fdt, const char *fdt_cmdline)
 	setprop_string(fdt, "/chosen", "bootargs", cmdline);
 }
 
+static int fdt_overrides_atag_mem(void *fdt)
+{
+	const char *memory;
+	int len = 0;
+
+	memory = getprop(fdt, "/memory", "reg", &len);
+	if (memory) {
+		while (len--) {
+			if (*memory != '\0')
+				return 1;
+		}
+	}
+
+	return 0;
+}
+
 /*
  * Convert and fold provided ATAGs into the provided FDT.
  *
@@ -180,7 +196,7 @@  int atags_to_fdt(void *atag_list, void *fdt, int total_space)
 		}
 	}
 
-	if (memcount) {
+	if (memcount && !fdt_overrides_atag_mem(fdt)) {
 		setprop(fdt, "/memory", "reg", mem_reg_property,
 			4 * memcount * memsize);
 	}